GenAI & Data Engineering Leader. Articles about data engineering, GenAI, distributed systems, and software architecture.https://kondasamy.com/en-ushttps://kondasamy.com/blog/2026/hermes-agent-personal-agent-postmortem/https://kondasamy.com/blog/2026/hermes-agent-personal-agent-postmortem/I run several coding agents every day. Hermes is different because its job is not to finish one coding task. Its job is to remember, schedule, and route the work that fills the rest of my day.Sun, 09 Aug 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/deslop-stack-fallacy-style-rules-hurt-thinking/https://kondasamy.com/blog/2026/deslop-stack-fallacy-style-rules-hurt-thinking/Developer Twitter claims you can deslop AI output by pasting ASD-STE100 or Orwell's rules into system prompts. Empirical benchmarks prove the opposite: forcing simple language at reasoning time cripples the model.Thu, 06 Aug 2026 00:00:00 GMThttps://kondasamy.com/til/2026/jsonl-for-ai-agent-workloads/https://kondasamy.com/til/2026/jsonl-for-ai-agent-workloads/Why JSONL shows up everywhere in agent tooling: streaming output, crash-safe logs, and fine-tuning datasets all need self-contained records that never wait for a closing bracket.Wed, 05 Aug 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/herdr-agent-multiplexer-gap/https://kondasamy.com/blog/2026/herdr-agent-multiplexer-gap/Once you run several coding agents at once, plain terminals and classic multiplexers stop being enough. Herdr fills the gap between tmux and agent apps.Sun, 02 Aug 2026 00:00:00 GMThttps://kondasamy.com/til/2026/pi-tree-vs-claude-resume/https://kondasamy.com/til/2026/pi-tree-vs-claude-resume/Pi's /tree navigates branches inside one session. Claude's /resume switches between saved conversations. Same word family, different jobs.Thu, 30 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/kimi-k3-open-frontier-systems-deep-dive/https://kondasamy.com/blog/2026/kimi-k3-open-frontier-systems-deep-dive/I spent a weekend with Moonshot's Kimi K3 launch notes and tech report. The useful story is not leaderboard rank. It is systems: sparse MoE, long context, open weights, and what it takes to run a 3T-class agent brain.Wed, 29 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/omp-subagent-system-deep-dive/https://kondasamy.com/blog/2026/omp-subagent-system-deep-dive/Most coding agents treat a subagent as a nested chat that returns prose. omp gives workers typed results, direct peer messages, and safer shared-file edits.Sun, 19 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/pi-agent-multi-tool-workflow/https://kondasamy.com/blog/2026/pi-agent-multi-tool-workflow/After 2 months switching between Claude Code, Codex, Gemini CLI, and Pi Agent for different coding tasks, Pi's 'primitives not features' philosophy turned it into the connective tissue between heavier tools.Tue, 14 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/tau-coding-agent-architecture-dissection/https://kondasamy.com/blog/2026/tau-coding-agent-architecture-dissection/HuggingFace's Tau is the first coding agent built to be understood. Its 200-line loop confirms what I suspected after 2+ years inside Claude Code, Cursor, and Codex: the architecture is universal, the magic is in the harness.Sun, 12 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/taste-skill-anti-slop-frontend-framework/https://kondasamy.com/blog/2026/taste-skill-anti-slop-frontend-framework/Every AI coding agent produces the same Tailwind + Inter + gradient hero. Taste Skill is an open-source framework that encodes design taste into portable SKILL.md files, turning generic UI slop into intentional, high-end frontends.Sat, 04 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/agentic-vs-agentive-accountability-gap/https://kondasamy.com/blog/2026/agentic-vs-agentive-accountability-gap/CMU's Critique of Agent Model gets the taxonomy right. Most AI agents are orchestrators, not autonomous systems. But 40+ pages pursuing true autonomy and barely a mention of institutional accountability reveals the field's real blind spot.Thu, 02 Jul 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/formal-verification-proof-oracles-ai-agents/https://kondasamy.com/blog/2026/formal-verification-proof-oracles-ai-agents/Tests sample. Formal verification proves. AI agents finally make these tools practical by writing specs at token cost and iterating on counterexamples for free.Tue, 30 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/mem0-postmortem-ai-memory-core-system/https://kondasamy.com/blog/2026/mem0-postmortem-ai-memory-core-system/I've built custom memory extraction for personalization and context compaction across multiple products. Here's what mem0 gets right, what's overcomplicated, and the three subsystems worth stealing if you're starting fresh.Mon, 29 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/ponytail-lazy-senior-dev-agent-governance/https://kondasamy.com/blog/2026/ponytail-lazy-senior-dev-agent-governance/Your AI coding agent isn't broken. It's over-eager. Ponytail forces agents through a six-level laziness ladder that cuts generated code by 80-94% while preserving the stuff that actually matters.Thu, 25 Jun 2026 00:00:00 GMThttps://kondasamy.com/til/2026/git-url-hackable-repo-tools/https://kondasamy.com/til/2026/git-url-hackable-repo-tools/Five tools that transform GitHub repos into AI-friendly formats, all using the same URL-substitution trick.Thu, 25 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/headroom-context-compression-agents/https://kondasamy.com/blog/2026/headroom-context-compression-agents/Your agent isn't slow or dumb. It's drowning in context it doesn't need. Headroom compresses 60-95% of tokens before they hit the model, and the model can pull back the originals on demand.Sat, 20 Jun 2026 00:00:00 GMThttps://kondasamy.com/til/2026/kami-constraint-based-document-design/https://kondasamy.com/til/2026/kami-constraint-based-document-design/A design system that makes AI agents produce print-quality documents by constraining, not expanding, the visual space.Wed, 17 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/html-diagram-skills-effective-html-archify/https://kondasamy.com/blog/2026/html-diagram-skills-effective-html-archify/Everyone writes 200-word prompts for architecture diagrams. One npx command installs a skill that produces better output every time.Sat, 13 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/adversarial-agents-anthropic-generator-evaluator/https://kondasamy.com/blog/2026/adversarial-agents-anthropic-generator-evaluator/Anthropic's internal harness for multi-hour agent runs borrows from GANs: a generator that builds, an evaluator that actually uses the app, and a contract they negotiate before any code is written.Tue, 09 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/generative-ui-missing-layer-agent-apps/https://kondasamy.com/blog/2026/generative-ui-missing-layer-agent-apps/We got reasoning, planning, and tool-calling. But nobody shipped the interface layer. That's why most agent apps still look like 2023 ChatGPT wrappers.Fri, 05 Jun 2026 00:00:00 GMThttps://kondasamy.com/til/2026/iterm2-scrollback-lines/https://kondasamy.com/til/2026/iterm2-scrollback-lines/The one setting I look up every time I set up a new Mac.Wed, 03 Jun 2026 00:00:00 GMThttps://kondasamy.com/til/2026/ccstatusline-claude-code-status-bar/https://kondasamy.com/til/2026/ccstatusline-claude-code-status-bar/A fully-featured TUI-configurable status line for Claude Code that makes my hand-rolled bash script look like a cave painting.Tue, 02 Jun 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/microvms-agent-sandboxing-regulated-environments/https://kondasamy.com/blog/2026/microvms-agent-sandboxing-regulated-environments/When your AI agent processes Patient A's records, can you guarantee it physically cannot touch Patient B's data? Containers can't. MicroVMs can.Sat, 30 May 2026 00:00:00 GMThttps://kondasamy.com/til/2026/understand-anything-codebase-knowledge-graph/https://kondasamy.com/til/2026/understand-anything-codebase-knowledge-graph/A Claude Code plugin that turns any codebase into an interactive knowledge graph you can explore, search, and ask questions about.Sun, 24 May 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/ai-native-patterns-software-3/https://kondasamy.com/blog/2026/ai-native-patterns-software-3/Most teams bolt LLMs onto existing architectures. AI-native teams delete the architecture entirely. Here are 10 patterns that show the difference.Fri, 22 May 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/claude-code-large-codebases-personal-playbook/https://kondasamy.com/blog/2026/claude-code-large-codebases-personal-playbook/Anthropic just published their enterprise playbook for Claude Code at scale. After deploying it across greenfield projects and legacy monoliths for over a year, here's what I'd add.Sun, 17 May 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/unreasonable-effectiveness-of-html/https://kondasamy.com/blog/2026/unreasonable-effectiveness-of-html/I needed a presentation. Every PPTX skill failed. Then I asked Claude Code to make an HTML file with a DESIGN.md, and had exactly what I wanted in minutes.Sat, 16 May 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/harness-engineering-reliable-ai-agents/https://kondasamy.com/blog/2026/harness-engineering-reliable-ai-agents/After months of AI agent failures across coding, customer support, and research workflows, I discovered the problem was never the model. An open-source course on harness engineering explained every failure mode.Sat, 09 May 2026 00:00:00 GMThttps://kondasamy.com/til/2026/macos-preview-pdf-editing/https://kondasamy.com/til/2026/macos-preview-pdf-editing/I spent years uploading PDFs to random websites for basic edits. Preview has been doing it all along.Wed, 29 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/design-md-consistent-ai-ui/https://kondasamy.com/blog/2026/design-md-consistent-ai-ui/AI agents can write React components on demand. What they can't do is remember what your product looks like. DESIGN.md fixes that with a single Markdown file your agent reads before it writes code.Fri, 24 Apr 2026 00:00:00 GMThttps://kondasamy.com/til/2026/mole-mac-storage-cleaner/https://kondasamy.com/til/2026/mole-mac-storage-cleaner/A single open-source binary that replaced CleanMyMac, AppCleaner, and DaisyDisk on my machine.Tue, 21 Apr 2026 00:00:00 GMThttps://kondasamy.com/til/2026/alphaxiv-paper-lookup-claude-code/https://kondasamy.com/til/2026/alphaxiv-paper-lookup-claude-code/Stop uploading raw PDFs to Claude. AlphaXiv serves pre-built markdown summaries of any arXiv paper for free.Mon, 20 Apr 2026 00:00:00 GMThttps://kondasamy.com/til/2026/portless-localhost-named-urls/https://kondasamy.com/til/2026/portless-localhost-named-urls/How Portless replaces fragile localhost:3000 ports with stable .localhost URLs that both humans and AI agents can rely on.Thu, 16 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/claude-code-source-leak-agent-builders/https://kondasamy.com/blog/2026/claude-code-source-leak-agent-builders/The biggest AI coding tool had its internals exposed. The surprise? No magic. Just 12 composable patterns that any developer can replicate.Wed, 15 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/15-patterns-production-ai-agents/https://kondasamy.com/blog/2026/15-patterns-production-ai-agents/Your agent demos work great. Then production happens. Here are the 15 resilience, containment, and observability patterns that separate toy agents from systems you can actually trust.Mon, 13 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/ai-avatar-platforms-compared/https://kondasamy.com/blog/2026/ai-avatar-platforms-compared/I tested four real-time AI avatar platforms for conversational agents. One renders micro-expressions at sub-1s latency. Another costs $0.009/min but looks like a Snapchat filter. Here's what matters when the avatar is the product.Fri, 10 Apr 2026 00:00:00 GMThttps://kondasamy.com/til/2026/install-docker-ubuntu-hetzner/https://kondasamy.com/til/2026/install-docker-ubuntu-hetzner/The complete Docker installation script I run every time I provision a fresh Hetzner server.Fri, 10 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/ai-design-copilots-compared/https://kondasamy.com/blog/2026/ai-design-copilots-compared/Every AI-generated UI looks the same. Purple gradients, Inter font, cards inside cards. Here's how Impeccable, Anthropic's frontend-design skill, Figma AI plugins, and prompt-to-app builders each try to fix that problem.Thu, 09 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/compound-engineering-speed/https://kondasamy.com/blog/2026/compound-engineering-speed/After a week with GSD, I switched to the Compound Engineering plugin. The difference isn't features. It's that your system gets faster, not just your chat.Tue, 07 Apr 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/livekit-vs-pipecat-voice-agents/https://kondasamy.com/blog/2026/livekit-vs-pipecat-voice-agents/I built real-time voice agents with both frameworks. LiveKit got me to a working call in 20 minutes. Pipecat took an afternoon of reading docs about pipelines, frames, and runners before I heard a voice.Tue, 31 Mar 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/gsd-vs-speckit-bmad-compound-engineering/https://kondasamy.com/blog/2026/gsd-vs-speckit-bmad-compound-engineering/Four frameworks all claim to fix vibe coding. They solve fundamentally different problems. Here's what each actually does and when to reach for which.Mon, 30 Mar 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/pros-dont-vibe-they-control/https://kondasamy.com/blog/2026/pros-dont-vibe-they-control/A UC San Diego study of 112 professional developers found the opposite of what Twitter preaches: experienced developers don't vibe code. They plan, prompt with surgical precision, and verify every line.Tue, 24 Mar 2026 00:00:00 GMThttps://kondasamy.com/til/2026/sips-macos-image-processing/https://kondasamy.com/til/2026/sips-macos-image-processing/The built-in macOS command that replaced my Preview/Photoshop habit for resizing, converting, and batch-processing images.Mon, 23 Mar 2026 00:00:00 GMThttps://kondasamy.com/til/2026/claude-code-tabs-vs-spaces-edit-failures/https://kondasamy.com/til/2026/claude-code-tabs-vs-spaces-edit-failures/Why Claude Code's Edit tool keeps failing on tab-indented codebases, and how to fix it.Mon, 16 Mar 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/grep-is-not-bm25/https://kondasamy.com/blog/2026/grep-is-not-bm25/I caught myself explaining BM25 as 'fancy grep' and realized that's fundamentally wrong. They solve different problems at different points on the search spectrum.Fri, 13 Mar 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/dense-vs-sparse-retrieval/https://kondasamy.com/blog/2026/dense-vs-sparse-retrieval/Everyone switched to vector search because embeddings feel like magic. But BM25 still wins on exact matches, rare entities, and domains where you can't afford to miss. Here's the real comparison.Thu, 12 Mar 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/multimodal-rag-patterns/https://kondasamy.com/blog/2026/multimodal-rag-patterns/Most teams reach for multimodal embeddings when a caption and a text pipeline would get them 80% of the way. Here are the three patterns for RAG beyond text, and when each one actually matters.Wed, 04 Mar 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/how-claude-code-works/https://kondasamy.com/blog/2026/how-claude-code-works/Most developers treat Claude Code like a chatbot with file access. It's not. It's a closed-loop agent where all intelligence lives in the model and all side effects live in the client.Sun, 15 Feb 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/workflow-vs-agent-vs-multi-agent/https://kondasamy.com/blog/2026/workflow-vs-agent-vs-multi-agent/Most 'agents' in production are workflows. Here's how to pick the right architecture without over-engineering or under-delivering.Tue, 03 Feb 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/what-is-an-ai-agent/https://kondasamy.com/blog/2026/what-is-an-ai-agent/Everyone's building AI agents, but nobody agrees on what one actually is. Here's what the pioneers, practitioners, and skeptics really mean when they say 'agent'.Thu, 29 Jan 2026 00:00:00 GMThttps://kondasamy.com/til/2026/intercom-s3-backup-empty-files/https://kondasamy.com/til/2026/intercom-s3-backup-empty-files/How I deleted 17,164 zero-byte files from S3 before discovering Intercom has a setting to prevent thisTue, 27 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/rag-for-dummies/https://kondasamy.com/blog/2026/rag-for-dummies/Everyone talks about RAG like it's magic. It's not. It's plumbing. Here's the practical guide to embeddings, vector databases, and retrieval that actually works in production.Mon, 19 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/uv-python-package-manager/https://kondasamy.com/blog/2026/uv-python-package-manager/After years of juggling pip, venv, pyenv, and pipx, UV consolidates everything into one tool that's 10-100x faster. Here's the complete cheat sheet.Sun, 18 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/modern-developer-tools-2026/https://kondasamy.com/blog/2026/modern-developer-tools-2026/The tools I use daily are 10-100x faster than what I used two years ago. Here's what changed and why.Fri, 16 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/claude-code-context-fundamentals/https://kondasamy.com/blog/2026/claude-code-context-fundamentals/Everyone's chasing flashy demos while missing the fundamentals. Here's what actually matters: treating your docs like infrastructure, making them machine-readable, and thinking out loud instead of typing compressed garbageThu, 15 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/browser-extension-frameworks-wxt/https://kondasamy.com/blog/2026/browser-extension-frameworks-wxt/GitHub stars don't ship features. WXT has emerged as the clear choice for browser extensions in 2026 while Plasmo's 8-month release gap signals real riskWed, 14 Jan 2026 00:00:00 GMThttps://kondasamy.com/til/2026/claude-code-custom-statusline/https://kondasamy.com/til/2026/claude-code-custom-statusline/How to build a single-line status bar that shows git context on the left and session metrics on the right.Tue, 13 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/postgres-explain-analyze-speed/https://kondasamy.com/blog/2026/postgres-explain-analyze-speed/Your team wants Redis. Your database needs indices. We got 97% faster with EXPLAIN ANALYZE instead of architectural rewrites.Sun, 11 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/modern-terminal-setup/https://kondasamy.com/blog/2026/modern-terminal-setup/After years of using default terminal tools, I rebuilt my workflow with modern alternatives. The result? 100x faster file searches, 80% fewer keystrokes, and workflows that don't require a mouse. Here's what actually matters.Mon, 05 Jan 2026 00:00:00 GMThttps://kondasamy.com/blog/2026/claude-code-interview-mode/https://kondasamy.com/blog/2026/claude-code-interview-mode/I discovered a game-changing approach to building features with Claude Code: interview first, spec second, code last. Here's how slowing down actually speeds things up.Fri, 02 Jan 2026 00:00:00 GMThttps://kondasamy.com/til/2025/git-worktree-ai-coding-agents/https://kondasamy.com/til/2025/git-worktree-ai-coding-agents/How git worktree enables parallel AI agent workflows - run multiple Claude Code or Cursor sessions simultaneously without context switching overhead.Wed, 31 Dec 2025 00:00:00 GMThttps://kondasamy.com/til/2025/git-worktree-ux-tools/https://kondasamy.com/til/2025/git-worktree-ux-tools/Three tools that fix git worktree UX so coding agents can run in parallel without setup pain.Wed, 31 Dec 2025 00:00:00 GMThttps://kondasamy.com/blog/2025/claude-code-tips-tricks/https://kondasamy.com/blog/2025/claude-code-tips-tricks/After 1.5 years with Cursor, Claude Code pulled me into its ecosystem. Here are the underrated features, keyboard shortcuts, and workflow patterns that made the difference.Tue, 30 Dec 2025 00:00:00 GMThttps://kondasamy.com/blog/2025/agentic-design-patterns/https://kondasamy.com/blog/2025/agentic-design-patterns/The most powerful AI isn't a single genius brain-it's teams of specialists arguing, reviewing their own work, and signing contracts. Here's how modern agentic systems actually work.Mon, 29 Dec 2025 00:00:00 GMThttps://kondasamy.com/blog/2025/engineering-voice-ai-latency-prompts/https://kondasamy.com/blog/2025/engineering-voice-ai-latency-prompts/Real engineering lessons from building Path AI at Layerpath - tackling sub-second latency, semantic drift in prompts, and why telling an LLM what NOT to do often backfires.Sat, 27 Dec 2025 00:00:00 GMThttps://kondasamy.com/til/2025/pnpm-outdated-upgrade-dependencies/https://kondasamy.com/til/2025/pnpm-outdated-upgrade-dependencies/Learn how to identify and safely upgrade outdated packages using pnpm's built-in commandsWed, 24 Dec 2025 00:00:00 GMThttps://kondasamy.com/blog/2025/blog-debugging-langfuse-500/https://kondasamy.com/blog/2025/blog-debugging-langfuse-500/Lessons learned from debugging K8s CrashLoopBackOff of self-hosted LangfuseTue, 23 Dec 2025 00:00:00 GMThttps://kondasamy.com/til/2025/setup-multiple-ssh-keys-macbook/https://kondasamy.com/til/2025/setup-multiple-ssh-keys-macbook/Manage multiple GitHub accounts with different SSH keys on macOS - seamless switching without conflicts.Mon, 15 Dec 2025 00:00:00 GMThttps://kondasamy.com/blog/2025/saas-boilerplate-nextjs/https://kondasamy.com/blog/2025/saas-boilerplate-nextjs/I was browsing the internet, looking at SaaS boilerplates. You know, those starter templates that promise to save you months of work.Fri, 12 Dec 2025 00:00:00 GMThttps://kondasamy.com/blog/2025/building-voice-ai-agents/https://kondasamy.com/blog/2025/building-voice-ai-agents/Voice AI isn't text-to-speech bolted onto a chatbot. The 300ms latency rule, real-time orchestration, and streaming architectures create fundamentally different engineering challenges.Tue, 09 Dec 2025 00:00:00 GMThttps://kondasamy.com/til/2025/apache-iceberg-time-travel/https://kondasamy.com/til/2025/apache-iceberg-time-travel/Query historical data by timestamp or snapshot ID using Iceberg's time travel syntax - a game-changer for debugging data pipelines.Tue, 09 Dec 2025 00:00:00 GMThttps://kondasamy.com/til/2025/rust-ownership-basics/https://kondasamy.com/til/2025/rust-ownership-basics/The three ownership rules and mental model that finally made Rust's borrow checker click.Tue, 09 Dec 2025 00:00:00 GMT